Re: NPAR TESTS

On Sunday 28 March 2004 22:07, you wrote:
> As I understand it, RANK is essentially a matter of making a copy
> of the active file and sorting the copy, then adding a variable
> to the active file that describes the resulting sort order.  If
> so, then it's probably a matter of augmenting our sort
> functionality.  This will take a little bit of work--I want to
> refactor the sort procedure anyway--but it shouldn't be a huge
> deal.
>
> Is the Wilcoxon signed-rank statistic important?  It's probably
> better to put it off until RANK is implemented.

Actually, there's several that need the rank.  W. S-R is a little more 
complicated than that even: you have to pool the scores from both variables, 
then assign the ranks and then put them back in the right variables.  There's 
a number of stats like that.

I'll leave rank to you, but I will need it for quite a number of tests.  I'll 
work on the simpler ones for now.

> Wouldn't surprise me if they could.  It's basically a sort, and
> external sorts are not difficult to write.  They probably just
> reuse the code from SORT CASES, the same way I'm planning to do
> for RANK.  It might take a little extra effort because you have
> to do *two* sorts--the first to get the ranks and the second to
> restore the same case ordering as in the active file--but
> certainly doable.

The added complication is that if you have 2 or more observations with the 
same rank, their rank is the weighted mean of the ranks they would otherwise 
be assigned.  That requires looking ahead in the sorted temp file or seeking 
back.

- M.